HERV2365 upregulates FGF1 expression by sponging miR-326 to promote the progression of intrahepatic cholangiocarcinoma

Summary

Intrahepatic cholangiocarcinoma (ICC) is an aggressive primary liver cancer with a dismal prognosis, underscoring the need to identify previously unrecognized molecular regulators involved in ICC progression. Although human endogenous retroviruses (HERVs) are aberrantly activated in multiple malignancies, their role in ICC remains poorly defined. Here, through integrative profiling of bulk and single-cell transcriptomes, we identified HERV2365 as a key upregulated oncogenic factor, which was enriched in ICC malignant cells and associated with poor prognosis. Functional assays demonstrated that HERV2365 enhanced ICC proliferation and tumorigenesis both in vitro and in vivo. Mechanistically, HERV2365 sponged miR-326, thereby derepressing fibroblast growth factor 1 (FGF1) and promoting tumor growth. Collectively, our findings identified HERV2365 as an unfavorable prognostic marker by promoting ICC progression through regulating the miR-326/FGF1 axis.

Graphical abstract

Highlights

• •HERV2365 is upregulated in ICC and enriched in malignant cells at the single-cell level
• •HERV2365 promotes ICC proliferation, migration/invasion, and tumor growth
• •HERV2365 serves as a ceRNA that sequesters tumor-suppressive miR-326 to derepress FGF1
• •The HERV2365/miR-326/FGF1 axis presents therapeutic opportunities in ICC

Hepatology; Molecular biology; Cancer

Introduction

Intrahepatic cholangiocarcinoma (ICC) is the second most prevalent primary hepatic cancer; it is derived from the epithelial lining of the intrahepatic bile ducts, and it accounts for approximately 10–15% of primary liver tumors worldwide.¹ During recent decades, both the incidence and mortality of ICC have continued to rise, with particularly high rates in East Asia and Southeast Asia, where genetic, environmental, and infectious risk factors are concentrated.^2,3 Recognized predisposing conditions include chronic liver fluke infection (Opisthorchis viverrini or Clonorchis sinensis), primary sclerosing cholangitis, hepatolithiasis, chronic viral hepatitis, and cirrhosis; however, a large proportion of patients present without identifiable causes.^4,5,6,7,8 Because ICC often progresses without symptoms and lacks sensitive diagnostic biomarkers, the majority of cases are detected at late stages when curative resection is no longer possible. Consequently, outcomes are poor, with a five-year overall survival (OS) of less than 10% and a median survival of under two years, even following surgery.^9,10,11 Although systemic chemotherapy with gemcitabine and cisplatin, as well as targeted and immunotherapeutic agents, has modestly improved prognosis, its efficacy remains inadequate, emphasizing the urgent need for novel molecular biomarkers and therapeutic strategies.¹²

Human endogenous retroviruses (HERVs), which are remnants of ancient retroviral infections integrated into the human germline, collectively constitute approximately 8% of the human genome.^13,14 Structurally, full-length HERVs resemble proviruses, containing viral genes (gag, pro, pol, env) flanked by long terminal repeats (LTRs) that harbor rich regulatory elements.¹⁵ Most HERV loci have lost their ability to encode proteins due to mutational inactivation and are usually silenced in normal cells by DNA methylation, histone modifications, and m6A modifications of RNA.^16,17,18,19 Nevertheless, accumulating evidence shows that HERVs can become aberrantly reactivated in pathological contexts, including the aging process,²⁰ autoimmune diseases,²¹ neurological disorders,^22,23 and malignancies.^24,25 In cancer, abnormal HERV expression is frequently observed, where it may act as promoter or enhancer, alter chromatin accessibility, generate noncoding transcripts, or modulate immune responses through virus-like proteins.^26,27,28

Among the functional outputs of HERV activation, HERV-derived long noncoding RNAs (lncRNAs) represent a significant transcriptional manifestation of HERV loci. These transcripts arise from HERV promoters or LTRs and can influence gene expression through multiple layers of regulation, including chromatin remodeling, transcriptional modulation, and post-transcriptional interactions. In cancer contexts, HERV-derived lncRNAs have been implicated in oncogenic processes such as cell proliferation, migration, and immune modulation, and in some cases function as competing endogenous RNAs (ceRNAs) that sequester microRNAs, thereby altering the expression of oncogenic or tumor suppressive targets.^29,30,31 However, the extent to which HERV-derived regulatory transcripts contribute to tumorigenesis varies across tumor types and remains an active area of investigation.

The potential role of HERV reactivation in ICC remains largely unexplored. Current research on ICC has focused mainly on genetic drivers such as IDH1/2 mutations, fibroblast growth factor receptors (FGFR2) fusions, alterations in chromatin-modifying genes, and classical tumor suppressors.^32,33 Whether HERV activation contributes to ICC development or progression represents a significant knowledge gap. Given the evolving recognition of HERV dysregulation across cancers and its potential to reshape transcriptional landscapes, it is critical to clarify whether HERV activation contributes to ICC progression.³⁴

In the present study, we systematically analyzed HERV expression in ICC using the ERVmap pipeline, which enables locus-specific quantification of endogenous retroviral transcription. Across multiple independent RNA-seq datasets, HERV2365 was identified as a consistently upregulated locus. Elevated HERV2365 expression was significantly associated with poor clinical outcomes. Functional assays demonstrated that HERV2365 enhanced ICC cell proliferation, migration, invasion, and cell-cycle progression. HERV2365 operates at a mechanistic level as a ceRNA by sponging and sequestering miR-326, which is known for its tumor-suppressive properties.³⁵ As a result, HERV2365 promoted the expression of fibroblast growth factor 1 (FGF1). Our research unveiled a novel regulatory axis involving HERV2365, miR-326, and FGF1 in ICC, which had not been previously recognized.

Results

Differential expression profile and prognostic role of HERVs in ICC

Using the ERVmap pipeline, we quantitatively analyzed RNA-seq data from ICC and adjacent non-tumor tissues, referencing a database of 3,220 HERV loci to achieve locus-specific mapping of RNA-seq reads and proviral sequences. In the GSE63420 dataset, 137 HERVs were significantly upregulated, and 155 were downregulated (Figure 1A). Analysis of GSE107943 identified 208 upregulated and 150 downregulated HERVs (Figure 1B), whereas GSE119336 revealed 42 upregulated and 79 downregulated loci (Figure 1C). Venn diagram intersection across all three datasets yielded 25 consistently upregulated and 51 consistently downregulated HERVs (Figures 1D and 1E; Table S2). The heatmap reveals a significant clustering between ICC and matched normal tissues, highlighting the transcriptional differences of HERV in ICC (Figures S1A–S1C).

Figure 1.

HERV dysregulation across ICC cohorts and construction of a HERV-based prognostic signature

(A–C) Volcano plots show differentially expressed HERVs in the GSE63420 (A), GSE107943 (B), and GSE119336 (C) datasets analyzed by the ERVmap pipeline.

(D and E) Venn diagrams illustrate overlapping up- (D) and down-regulated (E) HERVs across datasets.

(F and G) Coefficient profiles (F) and cross-validated partial-likelihood deviance (G) were used to select the optimal penalty.

(H) Forest plot of the LASSO regression model comprised 13 prognosis-related HERVs; hazard ratios (HRs) with 95% CIs were shown.

(I) Kaplan‒Meier curves for overall survival stratified by the risk score (GSE244807).

(J) Time‒dependent ROC curves for 1–, 3–, and 5–year survival prediction.

To investigate the potential significance of HERV expression in prognostic assessments, this research utilized the GSE244807 dataset, which includes survival data, to conduct LASSO regression analysis on 76 HERVs previously identified through differential analysis. Figures 1F and 1G present the coefficient curves and partial likelihood deviance, respectively. 10-fold cross-validation identified the optimal penalty parameter (λ) at which 13 HERVs retained non-zero coefficients, balancing model fit and complexity. A risk scoring model was subsequently developed on the basis of the expression levels of 13 HERVs along with their corresponding coefficient values. The forest plot indicated that, in addition to HERV565, which exhibited a protective effect, the remaining 12 HERVs were identified as risk factors (Figure 1H). This LASSO regression model facilitated the plotting of the Kaplan‒Meier survival curve related to OS (Figure 1I), demonstrating that patients in the high-risk category had significantly poorer prognoses compared to those in the low-risk group (p = 8.58e⁻¹³). To evaluate the predictive accuracy of this feature, transient ROC curve analysis was conducted. The results revealed AUC values of 0.784, 0.778, and 0.698 for predicting the 1–year, 3–year, and 5–year survival rates, respectively, indicating strong predictive capability (Figure 1J). High-risk patients exhibited increased infiltration of B cells, endothelial cells, eosinophils, M2 macrophages, mast cells, and regulatory T cells (Tregs), along with increased immune, stromal, and microenvironment scores (Figure S2A). These findings suggested that aberrant HERV expression may contribute to immunosuppressive microenvironments and result in poor prognosis in ICC.

HERV2365 is enriched in ICC malignant cells and associated with a poor prognosis, distinct expression pattern

Using scRNA-seq data from GSE138709, we performed stringent quality control and dimensionality reduction, annotating ten cell populations: dendritic cells, malignant cells, NK cells, T cells, macrophages, B cells, fibroblasts, endothelial cells, cholangiocytes, and hepatocytes (Figure 2A). Expression analysis of 13 prognosis-related HERVs was conducted. The results revealed that HERVs presented distinct expression patterns in different cell types, whereas HERV2365 was expressed mostly in malignant cells (Figures 2B, 2C, and S3A–S3I). Furthermore, we found that HERV2365 was significantly overexpressed in tumor tissues compared with adjacent normal tissues in the public cohorts (Figures 2D and 2E). Survival analysis revealed that high expression of HERV2365 was associated with poor prognosis in patients with ICC (Figure 2F). Using qRT-PCR technology, we confirmed its high expression in tumor tissues by analyzing 13 pairs of matched ICC samples (Figure 2G). Taken together, our results indicated that HERV2365 was enriched in malignant ICC cells, exhibited high expression levels in tumors, and showed a significant association with poor prognosis.

Figure 2.

HERV2365 is enriched in ICC and associated with poor prognosis

(A) t-SNE plots of 10 cell types identified by scRNA-seq (GSE138709).

(B and C) Feature plots (left diagram) and violin plots (right illustration) of normalized expression across the 10 cell types of the prognosis-related HERV loci, including HERV2365 (B), HERV565 (C).

(D and E) HERV2365 expression in ICC tissues from GSE107943 (D) and GSE63420 (E) datasets.

(F) Kaplan‒Meier survival analysis for HERV2365 (GSE244807).

(G) Relative expression levels of HERV2365 in 13 paired ICC and normal tissues by qRT-PCR. Data are presented as mean ± SD unless noted. Experiments were independently repeated ≥3 times; error bars denote SD. ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗p < 0.001.

HERV2365 drives malignant phenotypes in vitro and in vivo

Next, we evaluated the potential role of HERV2365 in ICC progression. Using RT-qPCR, we observed that the expression levels of HERV2365 in ICC cell lines (RBE, HCCC9810, HuCCT1, and SSP-25) were significantly higher than those in normal human intrahepatic biliary epithelial cells (HIBECs) (Figure 3A).

Figure 3.

HERV2365 promotes malignant phenotypes in vitro and in vivo

(A) Relative mRNA levels of HERV2365 in HIBEC, SSP-25, HCCC9810, HuCCT1, and RBE cell lines.

(B) siRNA knockdown efficiency by qRT-PCR in RBE and HCCC9810 cells.

(C) CCK-8 assays were performed in RBE and HCCC9810 cells transfected with HERV2365 siRNA.

(D) Colony formation assays of RBE and HCCC9810 cells.

(E) Wound healing assays were used to evaluate the migration of RBE and HCCC9810 cells, scale bars, 50 μm.

(F and G) Transwell migration (F), and invasion (G) assays were performed in RBE and HCCC9810 cells transfected with HERV2365 siRNA, scale bars, 50 or 100 μm.

(H) Flow cytometry shows the cell cycle distribution in RBE and HCCC9810 cells after HERV2365 knockdown.

(I) Representative images of subcutaneous xenograft tumors (n = 6).

(J and K) The weight and volume of subcutaneous xenograft tumors revealed the effects of HERV2365 knockdown on the formation of ICC. Data are presented as mean ± SD unless noted. Experiments were independently repeated ≥3 times; error bars denote SD. “ns” stands for “no significant difference,” ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, or ∗∗∗∗p < 0.0001.

To elucidate the biological function of HERV2365 in ICC, we employed siRNA to knock down the expression of HERV2365, and the knockdown efficiency was validated by qRT–PCR (Figure 3B). Initially, we assessed the impact of HERV2365 knockdown on the proliferation of ICC cells through CCK-8 assay and colony formation assays. The results demonstrated that the knockdown of HERV2365 inhibited the proliferation and growth capacity of RBE and HCCC9810 cells (Figures 3C and 3D). Scratch assay and transwell migration assays demonstrated that the knockdown of HERV2365 impeded cell migration, while the invasion assays indicated that the suppression of HERV2365 expression significantly reduced cell invasion capability (Figures 3E–3G). Flow cytometry was used to examine the cell cycle distribution in RBE and HCCC9810 cells post-transfection, and the results were compared between the RBE and HCCC9810 knockdown groups and the control group. The results revealed that upon HERV2365 knockdown, the proportion of cells in the G1 phase significantly increased, whereas the proportions in the S and G2 phases were markedly lower than those in the control cells. This outcome suggested that interference with HERV2365 led to a significant decrease in cell proliferation capacity (Figure 3H). Further studies revealed that in ICC cells with HERV2365 knockdown, the number of early apoptotic cells and late apoptotic cells showed a significant increasing trend, with the overall apoptosis rate also markedly elevated (Figure S4A). These data indicate that the reduction in HERV2365 expression level not only affects the cell cycle progression but also significantly impairs the anti-apoptotic function of the cells. Meanwhile, we overexpressed HERV2365 in RBE and HCCC9810 cells, and found that the proliferation and invasion abilities of ICC cells were significantly enhanced after HERV2365 overexpression (Figures S5A–S5C). These findings confirm the oncogenic role of HERV2365 in ICC cells.

To evaluate the in vivo role of HERV2365 in ICC, RBE cells stably transfected with sh-HERV2365 or control sh-NC were subcutaneously injected into nude mice. In the xenograft models, tumors derived from the HERV2365–knockdown group exhibited significantly reduced volumes and weights compared with those from the control group (Figures 3I–3K). These findings indicated that HERV2365 promotes ICC cell proliferation in vivo.

HERV2365 promotes ICC progression by regulating FGF1

ATAC-seq analysis of HIBEC, RBE, and HCCC9810 cells revealed that HERV2365 exhibited a pronounced chromatin opening state in RBE and HCCC9810 cells, indicating that HERV2365 is in an active transcriptional state in ICC (Figure 4A). To further elucidate the functional mechanisms of HERV2365, we selected candidate genes closely associated with tumor development for subsequent studies through RNA-seq (Figure 4B). Fibroblast growth factor 1 (FGF1) was identified as a key research target because of its critical role in regulating cell proliferation, migration, and angiogenesis through its interaction with FGFR, as well as its significantly downregulated expression pattern in the knockdown group. We therefore asked whether FGF1 mediates the effects of HERV2365 on tumor cell behavior. First, we found a positive correlation between HERV2365 and FGF1 in tumor tissues (Figures 4C and 4D). Second, knockdown of HERV2365 resulted in a significant decrease in FGF1 mRNA and protein expression levels in the RBE and HCCC9810 cell lines (Figures 4E and 4F).

Figure 4.

Association between HERV2365 and FGF1 expression in ICC

(A) ATAC-seq showed open chromatin at the HERV2365 locus in ICC cells.

(B) Heatmap of DEGs after HERV2365 knockdown in RBE cells via RNA-seq.

(C and D) Correlations between HERV2365 and FGF1 in GSE244807 (C) and 13 paired ICC tissues (D).

(E) FGF1 mRNA expression levels after HERV2365 knockdown in RBE and HCCC9810 cells.

(F) FGF1 protein levels after HERV2365 knockdown in RBE and HCCC9810 cells.

(G–I) FGF1 expression in the TCGA-CHOL (G), GSE107943 (H), and GSE63420 (I) datasets.

(J) The mRNA expression of FGF1 in 13 paired ICC tissues was measured by qRT-PCR.

(K) FGF1 relative expression in cell lines (RBE, HCCC9810 vs. HIBEC) by qRT-PCR.

(L) Kaplan-Meier analysis of FGF1 expression and overall survival in GSE244807. Data are presented as mean ± SD unless noted. Experiments were independently repeated ≥3 times; error bars denote SD. “ns” stands for “no significant difference,” ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, or ∗∗∗∗p < 0.0001.

Previous studies have demonstrated that FGF1 has a clear cancer-promoting effect, promoting tumor cell proliferation, angiogenesis, and metastasis through the FGFR signaling pathway, and is highly expressed in liver cancer, pancreatic cancer, colon cancer, and gastric cancer.³⁶ However, no studies have investigated its biological characteristics in ICC. Therefore, we first analyzed the expression of FGF1 in tumor tissues from public databases and collected ICC tissue samples, revealing that its expression was significantly greater in tumor tissues than in adjacent normal tissues (Figures 4G–4J). Additionally, high expression of FGF1 was observed in RBE cells, while no significant difference in FGF1 was found between HCCC9810 and HIBEC (Figure 4K). Survival analysis using the GSE244807 dataset indicated a correlation between FGF1 expression levels and patient prognosis, with patients exhibiting high FGF1 expression having poorer outcomes (Figure 4L).

Validation of FGF1 promoting malignant biological behaviors in ICC

To clarify the biological role of FGF1 in the development of ICC, we conducted a series of experiments involving the manipulation of FGF1 expression levels. First, since FGF1 is highly expressed in RBE cells, we utilized siRNA to effectively knock down the expression of FGF1 in RBE cells. Concurrently, given the low expression level of FGF1 in HCCC9810 cells, we achieved FGF1 overexpression in this cell line and confirmed the alterations in both mRNA and protein levels of FGF1 (Figures 5A and 5B). Subsequent analyses utilizing the CCK-8, scratch assay, and transwell migration assay revealed that the knockdown of FGF1 led to a marked reduction in both the proliferation and migration capabilities of RBE cells. In contrast, the overexpression of FGF1 in HCCC9810 cells was associated with a notable enhancement in ICC progression (Figures 5C–5E). Moreover, in vivo experiments demonstrated that the overexpression of FGF1 could promote tumor growth (Figures S6A–S6C). These results collectively confirmed the conclusion that FGF1 played a significant role in promoting the proliferation of ICC cells, underscoring its potential as an influential factor in the progression of this malignancy.

Figure 5.

FGF1 drives ICC growth and rescues functional loss following HERV2365 knockdown

(A) Validation of FGF1 knockdown in RBE cells at the mRNA and protein levels.

(B) qRT-PCR and western blot were used to detect the FGF1 overexpression in HCCC9810 cells.

(C) CCK-8 assay was performed in RBE transfected with FGF1 siRNA and in HCCC9810 transfected with FGF1 plasmid.

(D) Wound healing assays were used to evaluate the migratory ability of FGF1-knockdown RBE cells and FGF1-overexpressing HCCC9810 cells. Scale bars, 200 μm.

(E) Transwell migration assays were performed in RBE cells transfected with FGF1 siRNAs and in HCCC9810 cells with FGF1 plasmid, scale bars, 100 μm.

(F–H) Rescue experiments revealed that FGF1 overexpression restored proliferation and migration after HERV2365 knockdown in RBE and HCCC9810 cells, scale bars, 100 μm. Data are presented as mean ± SD unless noted. Experiments were independently repeated ≥3 times; error bars denote SD. “ns” stands for “no significant difference,” ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, or ∗∗∗∗p < 0.0001.

To further investigate whether HERV2365 promotes the proliferation and migration of ICC through FGF1 expression, we conducted rescue experiments involving single transfection of sh-HERV2365 and cotransfection of sh-HERV2365 and OE-FGF1 in RBE and HCCC9810 cells. When sh-HERV2365 and OE-FGF1 were cotransfected into RBE and HCCC9810 cells, the results of the CCK-8, scratch assays, and transwell migration assays demonstrated that FGF1 overexpression reversed the tumor-suppressive effects induced by sh-HERV2365 in ICC cells (Figures 5F–5H). We conducted a rescue experiment by transfecting siFGF1 into ICC cells overexpressing HERV2365, and the results demonstrated that knockdown of FGF1 could reverse the HERV2365 overexpression-induced promotion of ICC cell growth (Figures S7A and S7B). These findings suggest that HERV2365 promotes the development of ICC through the expression of FGF1.

Previous studies have demonstrated that FGF1 binds to the FGFR2 receptor, thereby inducing downstream ERK phosphorylation.³⁷ In our study, we knocked down FGF1 in RBE cells and performed western blot analysis to detect ERK and phosphorylated ERK (p-ERK). Our results revealed a significant decrease in p-ERK expression levels following FGF1 knockdown (Figure S8).

HERV2365 acts as a ceRNA to sponge miR-326 and derepress FGF1

To clarify the molecular nature of HERV2365, we first performed full-length annotation of the locus and transcript structure. HERV2365 spans 9,058 bp on chr6:144,923,390–144,932,448 (hg38) and is flanked by LTR elements, consistent with an endogenous retroviral origin (Figures S9A and S9B). Next, we assessed coding potential using four independent computational algorithms: CPAT, CNCI, PLEK2, and CPC2. Although CPC2 indicated a coding potential likely due to the transcript’s length sensitivity, leading to a higher likelihood of false positives, the other three algorithms consistently identified HERV2365 as non-coding. Specifically, the Coding Potential Assessment Tool (CPAT) yielded a probability of 0.174, well below the coding threshold of 0.364.³⁸ Similarly, both CNCI (score −0.036) and PLEK2 (score −0.876) classified the transcript as non-coding (Figure S9C). To further rule out the possibility of HERV2365 encoding a functional protein, we analyzed the longest putative open reading frame (ORF-1, corresponding to a 179-aa translated peptide, Table S3) for protein features. InterProScan analysis revealed that the putative ORF harbors no known functional domains or known motifs (0 matches) across major protein databases (Pfam, SMART, PROSITE, and so forth) (Figure S10A). Furthermore, BLAST/CDD detected no putative conserved domains, and BLASTP against the curated RefSeq Select protein database (refseq_select_prot) yielded no significant similarity (Figures S10B and S10C). To functionally validate these in silico predictions, we performed gain-of-function experiments by overexpressing ORF-1. Although ORF-1 was confirmed to be expressed, its overexpression did not elicit any significant changes in tumor-related phenotypes, including proliferation, migration, or invasion, in our experimental models. This absence of phenotypic effect suggests that, despite possessing translational potential, ORF-1 does not function as a protein driver of oncogenic processes in this context (Figures S11A–S11E). Together, these multi-dimensional in silico and in vitro analyses confirm that HERV2365 does not encode a functional protein and support that HERV2365 primarily acts as a non-coding transcript.

To elucidate how HERV2365 regulates the expression of FGF1, we employed FISH assays and found that the HERV2365 transcripts are predominantly localized in the cytoplasm of RBE and HCCC9810 cells (Figure 6A), which is consistent with a ceRNA-compatible subcellular distribution. We hypothesized that HERV2365 might regulate the expression of target genes by competitively binding to miRNAs, thereby relieving the inhibitory effects of miRNAs on downstream target genes. Consequently, we identified a candidate miRNA (miR-326) by cross-referencing miRNA target prediction results from the miRDB and the miRTarBase (Figure 6B). AGO2-RIP followed by RT-qPCR revealed that both HERV2365 and miR-326 were significantly enriched in the AGO2 immunoprecipitates compared with the IgG controls in RBE and HCCC9810 cells, indicating their association with AGO2-containing RISC complexes. These data support the notion that HERV2365 engages in miR-326–dependent post-transcriptional regulation, consistent with a ceRNA mechanism (Figure S12A). To validate the specific binding, we constructed a mutant plasmid of HERV2365 (HERV2365-MUT) by introducing site-directed mutations specifically into the predicted miR-326 binding region, changing the sequence from 5′-CCCAGAGC-3′ to 5′-GGGTCTCC-3' (as provided in Data S1A and Figure 6C). The dual-luciferase reporter assay demonstrated that cotransfection with the miR-326 mimics significantly reduced the luciferase activity of the wild-type HERV2365 reporter compared to the control group. Notably, this inhibitory effect was abolished when the binding site was mutated, thereby confirming that miR-326 directly targets HERV2365 at the predicted site (Figure 6C). Furthermore, the expression level of miR-326 was significantly increased after the knockdown of HERV2365 (Figure 6D), indicating that HERV2365 functions as a molecular sponge for miR-326 in ICC.

Figure 6.

HERV2365 functions as a ceRNA by sponging miR-326

(A) FISH shows the cytoplasmic localization of HERV2365 in RBE and HCCC9810 cells, scale bars, 100 μm.

(B) Candidate miRNA selection by miRDB/miRTarBase identified miR-326.

(C) The predicted miR-326 binding site (highlighted in red) within the HERV2365 wild-type (WT) sequence was mutated (MUT) to verify specificity, and the dual-luciferase assays were used to confirm the interaction between HERV2365 and miR-326.

(D) HERV2365 knockdown increased miR-326 expression in RBE and HCCC9810 cells.

(E) The expression of miR-326 in ICC tissues (TCGA-CHOL).

(F) The expression of miR-326 in RBE, HCCC9810, and HIBEC cells was measured via qRT-PCR.

(G) qRT-PCR was used to confirm the transfection of miR-326 mimics in RBE and HCCC9810 cells.

(H–J) CCK-8 (H), Wound-healing (I), and Transwell migration assays (J) show the inhibition of proliferation and migration by miR-326 mimics, scale bars, 100 μm.

(K and L) Rescue assays were applied to show that the miR-326 inhibitor reversed the effects of HERV2365 knockdown. Data are presented as mean ± SD unless noted. Experiments were independently repeated ≥3 times; error bars denote SD. “ns” stands for “no significant difference,” ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, or ∗∗∗∗p < 0.0001.

Studies have shown that miR-326 exerts a tumor-suppressive effect on many tumors by targeting NOTCH1, SMO, Twist1, and others, thereby inhibiting tumor cell proliferation, invasion, and stem cell characteristics.³⁹ However, its role in ICC remains unknown, and the literature suggested that miR-326 was expressed at low levels in cholangiocarcinoma tissues.⁴⁰ To investigate the role of miR-326 in the growth of ICC, we first analyzed its expression in the TCGA-CHOL cohort and found that miR-326 is expressed at low levels in tumor tissues. Compared with that in HIBEC, its expression was also lower in RBE and HCCC9810 cells (Figure 6E and 6F). Therefore, we transfected miR-326 mimics into these two cell lines and validated the results using qRT-PCR (Figure 6G). The results demonstrated that the miR-326 mimics significantly inhibited the proliferation and migration capabilities of RBE and HCCC9810 cells (Figures 6H–6J). Importantly, these findings were recapitulated in our in vivo experiments, which demonstrated that miR-326 significantly suppresses ICC tumorigenesis (Figures S13A–S13D). We subsequently confirmed through a rescue experiment that the miR-326 inhibitor reversed the inhibitory effects on ICC cells induced by sh-HERV2365 (Figures 6K and 6L). Conversely, the miR-326 mimics counteracted the growth-promoting effect of OE-HERV2365 on ICC cells (Figures S14A and S14B). Our findings confirmed that HERV2365 can regulate the proliferation and migration abilities of ICC cells through miR-326.

To validate that FGF1 is the downstream target gene of miR-326 in ICC, we first observed a negative correlation between the expression of FGF1 in tumor samples and that of miR-326 in the TCGA-CHOL cohort (Figure 7A). Using a luciferase reporter assay, we subsequently designed an FGF1 mutant plasmid (as provided in Data S1B and Figure 7B) on the basis of the binding region of FGF1 and miR-326. The results of the dual-luciferase reporter assay confirmed that FGF1 and miR-326 can directly and specifically bind (Figure 7B). Furthermore, we found that the mRNA and protein expression levels of FGF1 were also reduced in RBE and HCCC9810 cells transfected with miR-326 mimics (Figures 7C and 7D). Moreover, the miR-326 inhibitor increased the protein expression level of FGF1 after HERV2365 knockdown (Figure 7E). The above experiments demonstrate that HERV2365 can regulate the proliferation and migration of ICC cells through the miR-326/FGF1 axis.

Figure 7.

miR-326 directly targets FGF1 to mediate the HERV2365 axis

(A) Negative correlation between miR-326 and FGF1 expression in the TCGA-CHOL cohort.

(B) The predicted binding sequences of miR-326 within FGF1 3′-UTR (Wild-type, WT) were aligned with their respective mutant designs (Mutant, MUT), with mutated nucleotides highlighted in red, and the dual-luciferase assays were applied to validate the binding between miR-326 and the FGF1 3′-UTR.

(C and D) qRT-PCR and western blotting revealed that miR-326 mimics reduced FGF1 expression in RBE and HCCC9810 cells.

(E) Western blotting was used to analyze the expression of FGF1 in RBE and HCCC9810 cells cotransfected with HERV2365 knockdown and miR-326 inhibitor.

(F) Schematic model illustrates that HERV2365 promotes ICC by sponging miR-326 to derepress FGF1. Data are presented as mean ± SD unless noted. Experiments were independently repeated ≥3 times; error bars denote SD. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, or ∗∗∗∗p < 0.0001.

Discussion

ICC is a highly lethal malignancy of the liver that presents formidable clinical challenges. Despite advances in diagnostic imaging, surgical procedures, and systemic therapies, long-term survival rates remain poor, with the majority of patients diagnosed at advanced stages and relapse being common even after resection.^41,42 This underscores the urgency of uncovering the molecular mechanisms that drive ICC development and progression and of identifying reliable biomarkers and therapeutic targets. In this study, our data indicate that HERV2365 functions as a ceRNA, driving ICC progression by sequestering miR-326 to relieve its suppression of FGF1 (Figure 7F).

Although HERVs are considered to be genomic remnants without function, they are increasingly recognized as biologically active elements with potential roles in tumor biology.⁴³ Aberrant reactivation of HERVs has been reported in multiple cancers, where they may act as alternative promoters or enhancers, influence chromatin remodeling, or give rise to noncoding transcripts with oncogenic potential.^44,45 Given the emerging links between HERV activity and tumor immunity, we further explored the immune contexture by stratifying cases according to the global HERV signature. The HERV-high group displayed higher ESTIMATE-derived immune, stromal, and microenvironment scores and was accompanied by increased infiltration of immunosuppressive populations, including , Th2 cells, and M2 macrophages, consistent with the typically “cold” and stroma-rich immune milieu of ICC.^46,47,48 Mechanistically, derepressed HERV transcripts can trigger innate nucleic-acid sensing and “viral mimicry” programs that remodel type I interferon signaling and downstream checkpoint exhaustion states, while certain HERV Env proteins harbor immunosuppressive domains capable of dampening T cell activation, providing a plausible framework by which HERV2365 may contribute to immune evasion within a broader HERV-active landscape.^49,50,51

Building upon this broader HERV-active landscape, our identification of the HERV2365/FGF1 axis provides a crucial and specific mechanistic explanation for these infiltration patterns. Aberrant FGF/FGFR signaling has been widely implicated in tumor immune evasion and remodeling of the tumor microenvironment. While the direct immunological role of tumor-derived FGF1 remains to be fully clarified, FGF1 is a potent paracrine activator of FGFRs. Previous studies have suggested that hyperactive FGF/FGFR signaling may facilitate macrophage recruitment and promote their polarization toward a pro-tumorigenic, immunosuppressive M2-like phenotype.^52,53,54 In addition, aberrant FGF/FGFR cascades have been associated with Treg accumulation and immune checkpoint upregulation, including PD-L1, thereby contributing to the suppression of cytotoxic T cell function.^55,56 Therefore, we propose that HERV2365-mediated upregulation of FGF1 may serve as a plausible mediator that, together with broader HERV-associated mechanisms, contributes to remodeling of the tumor microenvironment and the establishment of an immunosuppressive niche that favors tumor progression.

Within the ceRNA framework, HERV-derived lncRNAs can act as molecular sponges that sequester tumor-suppressive microRNAs (miRNAs), thereby relieving the repression of oncogenes.⁵⁷ HULC enhances hepatocellular carcinoma growth by acting as a sponge for miR-2052⁵⁸; HOST2 promotes ovarian cancer cell migration through the sequestration of miR-let-7b⁵⁹; and linc-ROR fosters breast cancer progression by binding to miR-205 and facilitating epithelial-mesenchymal transition (EMT).⁶⁰ Our findings extend these observations to ICC by showing that HERV2365 is significantly overexpressed in tumor tissues and ICC cells and that its expression is correlated with cell migration and cell proliferation.

Functionally, we found that HERV2365 augmented ICC cell proliferation, clonogenic capacity and motility in vitro. Cell cycle analysis revealed that the depletion of HERV2365 induced G1–phase arrest, thereby impairing progression into S phase. Concomitantly, the apoptosis rate was significantly increased. These results suggest that the knockdown of HERV2365 may exert an anticancer effect by inhibiting cell cycle progression and promoting apoptosis. Mechanistically, we demonstrated that HERV2365 acted as a molecular sponge for miR-326, relieving its repression of FGF1. This conclusion was supported by luciferase assays confirming direct binding between HERV2365 and miR-326. Previous reports have established miR-326 as a tumor suppressor in various cancers, including glioma,⁶¹ gastric cancer,⁶² and pancreatic carcinoma.⁶³ Consistent with these findings, our research demonstrated that miR-326 was downregulated in ICC. Restoring its expression suppressed the malignant phenotype, and knocking down its expression reversed the inhibitory effect of HERV2365 silencing. As the principal effector in this circuit, FGF1 is a prototypic fibroblast growth factor that activates the FGFR-MAPK/ERK and PI3K/AKT cascades to promote proliferation, survival, angiogenesis, and motility.^64,65 Clinically relevant overexpression of FGF1 has been documented in gastrointestinal and other solid tumors and is correlated with an unfavorable prognosis.^66,67 In ICC, we observed increased FGF1 expression in tumor tissues and confirmed its association with unfavorable clinical outcomes. The observation that endogenous FGF1 was significantly upregulated in RBE cells while remaining unchanged in HCCC9810 cells underscores the biological heterogeneity of ICC. This expression disparity dictated our experimental strategy: We targeted the FGF1-high RBE cells for knockdown assays to evaluate the necessity of FGF1 in maintaining the malignant phenotype, while utilizing the FGF1-low HCCC9810 cells for overexpression assays to assess its sufficiency. Consequently, these distinct cellular baselines did not interfere with the reliability of the results but rather strengthened the conclusion that FGF1 functions as a critical oncogenic driver in ICC regardless of the initial expression context. Functional assays revealed that knockdown of FGF1 diminished proliferation and migration, whereas its ectopic expression enhanced malignant behavior. Rescue experiments further demonstrated that the overexpression of FGF1 counteracted the suppressive effects of HERV2365 silencing and knockdown of FGF1 was able to reverse the promoting effect of HERV2365 overexpression, validating the central role of the HERV2365/miR-326/FGF1 axis in ICC progression. Importantly, aberrant FGF/FGFR signaling is increasingly recognized as a key oncogenic axis in cholangiocarcinoma, and ligand-driven activation can cooperate with genetic FGFR alterations to sustain MAPK/ERK and PI3K/AKT outputs that are central to malignant transformation and tumor maintenance.^68,69 Consistent with this concept, we also found that FGF1 can activate the phosphorylation level of downstream ERK. Together, these findings suggest that FGF1 may contribute more broadly to ICC carcinogenesis by reinforcing autocrine/paracrine FGFR signaling programs that favor proliferation, survival, and pro-tumorigenic microenvironmental interactions.

In addition to its biological importance, this regulatory pathway carries translational significance. Both HERV2365 and FGF1 expression were associated with poor prognosis, suggesting their potential use as prognostic biomarkers. Moreover, given the rapid development of RNA-targeting therapeutic strategies, including antisense oligonucleotides, small interfering RNAs, and CRISPR-based genome editing, direct targeting of HERV2365 or modulation of its ceRNA activity could represent a promising therapeutic strategy.^70,71,72,73 Intervening in ceRNA networks more broadly may provide opportunities for novel treatment approaches in ICC, a cancer type that remains refractory to most current therapies.

In summary, we demonstrate that HERV2365 promotes ICC tumor progression by acting as a sponge for miR-326 and thereby upregulating FGF1. These findings advance our understanding of HERV-driven mechanisms in cancer and highlight HERV2365 as both a prognostic biomarker and a potential therapeutic target in ICC.

Limitations of the study

Although this study has made initial progress in understanding the mechanisms involved, several limitations regarding clinical relevance, the depth of clinical verification, and the complexity of the molecular mechanisms should be acknowledged.

First, the clinical relevance of our findings requires further validation. Because ICC is a highly aggressive malignancy and our study was constrained by strict inclusion criteria and the retrospective nature of sample collection, the internal validation cohort was limited to 13 paired tumors and adjacent tissues, with incomplete clinicopathological annotations and no longitudinal follow-up data. As a result, we were unable to perform robust analyses correlating HERV2365 or FGF1 expression with clinicopathological parameters or patient outcomes in our in-house cohort. Although prognostic analyses were supported by publicly available datasets, future studies using larger, independent, and ideally prospective cohorts with comprehensive clinical annotation will be essential to establish the translational and prognostic value of the HERV2365/miR-326/FGF1 axis in ICC.

Second, the depth of clinical verification in this study remains limited. Due to the restricted availability of biobank-derived tissue specimens, we were unable to perform immunohistochemical staining or other spatially resolved validation assays, and therefore could not directly assess the relationship between HERV2365 expression and FGF1 protein abundance in patient tissues. Additional validation at both the RNA and protein levels in expanded clinical cohorts will be important to further strengthen the clinical significance of our mechanistic observations.

Third, regarding the inherent complexity of the ceRNA regulatory network, we maintain an open attitude toward the multifaceted role of HERV2365. Although our rescue experiments strongly support the existence of a HERV2365/miR-326/FGF1 regulatory axis, we recognize that the ceRNA network is inherently complex, and our current data do not completely exclude the possibility that HERV2365 may also interact with additional miRNAs or regulate other downstream targets. Future studies should systematically characterize the broader HERV2365-associated ceRNA network and determine whether additional miRNAs, downstream effectors, or context-dependent interactions contribute to the observed phenotype.

Finally, although our study mainly focused on defining the HERV2365/miR-326/FGF1 axis, we also found that FGF1 increased the phosphorylation level of downstream ERK, providing preliminary evidence that this axis may be linked to the activation of oncogenic signaling pathways. However, the extension of the molecular mechanism remains limited, as the downstream signaling consequences were not systematically investigated in the current study. Future studies combining more comprehensive pathway analyses and functional assays will be needed to better define the downstream molecular landscape of HERV2365 in ICC.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Jianmin Wang (wangjm8605@163.com).

Materials availability

All experimental reagents were either commercially available or provided by collaborators under appropriate agreements, with all sources properly documented.

Data and code availability

• •The public datasets analyzed (GEO/TCGA) are listed in the manuscript.
• •The RNA-seq data generated in this study have been deposited in the NCBI Sequence Read Archive (SRA) under BioProject accession number PRJNA1336329 and are publicly available as of the date of publication.

  The data can be accessed at: https://www.ncbi.nlm.nih.gov/sra/PRJNA1336329

• •Western blot data: All original, uncropped western blot images, along with a descriptive file, are available as supplemental information (Data S1C).
• •Additional information: Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon reasonable request.
• •This paper does not report the original code.

Acknowledgments

The authors acknowledge the open-source data from TCGA and GEO and the Biobank of Fujian Cancer Hospital for providing the ICC samples. This research was supported by the Major Scientific Research Program for Young and Middle-aged Health Professionals of Fujian Province, China (grant no. 2021ZQNZD009), the research projects of Fujian Provincial Cancer Hospital (F2125Y-GSP01-01, F2326Y-YZK04-01 and F2326R-CXZX01-01), and the Science and Technology Innovation Joint Fund project of Fujian Province (no. 2025Y9661).

Author contributions

Y.Z.: conceptualization, data curation, formal analysis, investigation, visualization, methodology, validation, writing – original draft, and writing – review and editing. X.J.: supervision, software, methodology, and writing – review and editing. F.W.: visualization, supervision, methodology, validation, resources, and writing – review and editing. J.X.: visualization, supervision, and validation. Q.F.: investigation, resources, and methodology. G.Ye: methodology, resources, and investigation. Z.Gao: conceptualization, project administration, supervision, and writing – review and editing. J.W.: conceptualization, supervision, investigation, project administration, funding acquisition, and writing – review and editing. J.L.: conceptualization, supervision, investigation, project administration, and writing – review and editing. All the authors read and approved the final manuscript.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Experimental models: cell lines
HuCCT1	Procell (Wuhan, China)	Cat.CL-0725;RRID:CVCL_0324
HCCC9810	Procell (Wuhan, China)	Cat.CL-0095;RRID:CVCL_6908
RBE	Procell (Wuhan, China)	Cat.CL-0191;RRID:CVCL_4896
SSP-25	Cell Bank, Chinese Academy of Sciences (Shanghai, China)	CSTR:19375.09.3101HUMTCHu255;RRID:CVCL_4902
HEK293T	Cell Bank, Chinese Academy of Sciences (Shanghai, China)	CSTR:19375.09.3101HUMGNHu44; RRID:CVCL_0063
HIBEC	Zhejiang Meisen (Hangzhou, China)	CTCC-033-HUM
Experimental models: organisms/strains
BALB/c nude mice (male, 4 weeks)	SPF (Suzhou) Biotechnology	N/A
Oligonucleotides
siRNAs; miR-326 mimics/inhibitors	Shanghai Bioengineering	Sequences in Table S1
qPCR primers	This study	Sequences in Table S1
Antibodies
Anti-FGF1	Abcam	ab207321;RRID:AB_941577
Anti-GAPDH	Proteintech	60004-1-Ig;RRID:AB_2721282
Chemicals, peptides, and recombinant proteins
Polybrene	Beyotime	1 mg/mL (as used)
Puromycin	Beyotime	2 μg/mL (as used)
TRIzol reagent	Vazyme	N/A
RIPA buffer; PMSF	Beyotime	N/A
Critical commercial assays
FastPure Cell/Tissue Total RNA Isolation Kit V2	Vazyme	Cat.RC112-01
HiScript II Q RT SuperMix for qPCR (+gDNA wiper)	Vazyme	Cat.R223-01
Taq Pro Universal SYBR qPCR Master Mix	Vazyme	Cat.Q712-02
Dual-Luciferase Reporter System Kit	Vazyme	Cat.DL101-01
CCK-8	Yeason	Cat.40203ES76
PI staining solution	Yeason	Cat.40755ES64
Fluorescent In Situ Hybridization Kit	Servicebio	N/A
Deposited data
Bulk RNA-seq raw data	NCBI SRA	PRJNA1336329
Public ICC RNA-seq/clinical datasets	GEO	GSE107943; GSE63420; GSE119336; GSE244807
Public scRNA-seq dataset	GEO	GSE138709
TCGA cholangiocarcinoma cohort	TCGA	TCGA-CHOL
Software and algorithms
ERVmap	GitHub	https://github.com/mtokuyama/ERVmap
DESeq2	R package	N/A
TBtools	GitHub	https://github.com/CJ-Chen/TBtools
Cell Ranger	10x Genomics	v8.0.1
scTE	GitHub	https://github.com/JiekaiLab/scTE
Seurat; SeuratDisk	R packages	N/A
TIMER (xCell)	Web tool	http://timer.cistrome.org
survival; survminer	R packages	v3.5.8
timeROC	R package	N/A
HTSeq	Software	v0.6.1
GraphPad Prism	Software	v9.3.5

Experimental model and study participant details

Human participants

13 paired ICC tumor samples and matched adjacent non-tumor tissues were collected from patients undergoing radical resection at Fujian Cancer Hospital. Ethical approval for this study was granted by the Fujian Cancer Hospital Ethics Committee (K2024-307-01), and written informed consent was obtained from all participants.

Cell lines

The ICC cell lines (HuCCT1, HCCC9810, and RBE) were all purchased from Procell Life Science & Technology Co., Ltd.(Wuhan, China). The SSP–25 cell line and human embryonic kidney 293T cells (293T) were obtained from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). Additionally, normal human intrahepatic biliary epithelial cells (HIBEC) provided by Zhejiang Meisen Cell Technology Co., Ltd.(Hangzhou, China) were also used in this study. The HIBEC cells were cultured in primary cell culture medium, which was also purchased from Zhejiang Meisen Cell Technology Co., Ltd. All ICC cell lines were cultured in RPMI 1640 medium supplemented with 10% fetal bovine serum and a 1% penicillin‒streptomycin antibiotic mixture, and maintained at 37 °C in a 5% CO₂ environment. The cell lines used in this study were authenticated by short tandem repeat (STR) profiling to confirm their identity and genetic consistency. All experiments were performed with mycoplasma–free cells.

Animals

Four-week-old male BALB/c nude mice were obtained from SPF (Suzhou) Biotechnology and housed under specific pathogen-free conditions. All animal procedures were approved by the Ethics Committee of Fujian Cancer Hospital and complied with AAALAC and IACUC guidelines (FJMU 2025-Y-3207).

Method details

Public data collection

RNA sequencing (RNA–seq) data for ICC were retrieved from the NCBI Sequence Read Archive (SRA, https://www.ncbi.nlm.nih.gov/sra). Differential expression of human endogenous retroviruses (HERVs) was analyzed using datasets GSE107943, GSE63420, and GSE119336, while dataset GSE244807 was utilized for the construction of prognostic models. Single–cell transcriptome analysis was conducted using dataset GSE138709, and the TCGA–CHOL database was also included in the analysis.

ERVmap pipeline and differential expression analysis

The expression of full–length proviral sequences was quantified using the ERVmap pipeline (https://github.com/mtokuyama/ERVmap).³⁶ Differentially expressed HERVs were identified in GSE107943, GSE63420, and GSE119336 using DESeq2, with thresholds set at |log2FC| > 1.0 and adjusted p < 0.05. Heatmaps depicting expression patterns were generated using Tbtools (https://github.com/CJ–Chen/TBtools).

scRNA-seq and immune infiltration analysis

BAM alignment files were generated using the cellranger software (version 8.0.1). Subsequently, the scTE analysis pipeline (https://github.com/JiekaiLab/scTE) was employed to quantify the single–cell expression levels of HERVs, with gene annotation files sourced from GENCODE GRCh38.p14 and HERV annotation files obtained from the ERVmap database. The scTE quantification results were imported into the R environment using the SeuratDisk package, and Seurat objects from each sample were integrated. Data visualization was performed using the RunTSNE function to execute the t–distributed stochastic neighbor embedding (tSNE) technique. The identity of each cell cluster was characterized through ICC signature genes.³⁷ To further analyze the association between patient risk assessment and immune cell infiltration within the tumor microenvironment, we applied the xCell algorithm from the TIMER online database (http://timer.cistrome.org) to perform quantitative analysis of immune cell infiltration and immune scores in the GSE244807 dataset.

Construction of the HERV prediction model

This study initially employed univariate Cox regression analysis to identify survival–related genes, followed by LASSO regression analysis aimed at identifying potential prognostic biomarkers and constructing a prognostic scoring model associated with HERVs. The construction of this model was based on the expression levels of selected HERVs and their corresponding regression coefficients. This method allows for a quantitative analysis of the contribution of HERVs to the model’s predictive performance. Patients were divided into high–risk and low–risk groups on the basis of the median prognostic risk score. With the “survival” and “survminer” packages (version 3.5.8) in R, Kaplan‒Meier (K‒M) survival curves were plotted to test the significance of differences in overall survival (OS) between different risk groups. The validity and accuracy of the model were evaluated by calculating the area under the curve (AUC) at 1–year, 3–years, and 5–years intervals, and ROC curves were plotted using the “timeROC” package.

RNA extraction and quantitative real-time PCR

A FastPure Cell/Tissue Total RNA Isolation Kit V2 (Vazyme, China) was used to extract RNA from tissues and cells following the manufacturer’s instructions. Subsequently, the HiScript II Q RT SuperMix for qPCR (+gDNA wiper) (Vazyme, China) was subsequently used to reverse transcribe RNA into cDNA. Furthermore, we performed the reactions via the Taq Pro Universal SYBR qPCR Master Mix Kit (Vazyme, China) on the StepOnePlus Real–Time PCR System (Applied Biosystems). The relative expression levels were normalized to those of GAPDH or U6 as internal controls. The primer sequences are listed in Table S1.

Cell transfection

Small interfering RNAs (siRNAs), miR–326 mimics and inhibitors were synthesized by Shanghai Bioengineering Co., Ltd. According to the manufacturer’s instructions, Opti–MEM medium was used with Lipo3000proMAX cell transfection reagent (liposome) (Heyuan Biotechnology, China). The specific sequences of siRNAs, mimics, and inhibitors are listed in Table S1. To establish stably transfected cell lines, cells were transfected with lentiviruses expressing the negative control, sh–HERV2365, OE-miR-326, or OE–FGF1. sh–HERV2365, OE–FGF1, OE-miR-326, and polybrene (1 mg/mL, Beyotime, China) were cotransfected at a multiplicity of infection (MOI) of 20. After 72 h of transfection, the cells were treated with 2 μg/mL puromycin for 7 days to screen for stable cell lines.

CCK-8 proliferation assay

After the transfection procedure was complete, 2,000 cells were seeded per well. The cells were resuspended in 100 μL of culture medium, mixed thoroughly, and then seeded into a 96–well plate. At time points of 0 hours, 24 hours, 48 hours, 72 hours, and 96 hours, add 100 μL of culture medium containing 10 μL of CCK–8 reagent (Yeason, China) was added to each well. After two or three hours of incubation, measure the absorbance of each well at a wavelength of 450 nm using a microplate reader.

Wound healing assay

RBE and HCCC9810 cells were seeded into prelabeled 6–well plates. Once the cells reached full confluency, a straight wound was created on the cell surface by scratching with a 200 μL pipette tip. At 0 h, 12 h, and 24 h, microscopic photographs of the scratches in each well were taken. The scratch area was quantitatively analyzed via ImageJ software, and the cell migration index was calculated according to the following formula:migration index = (0–hour scratch area–12–hour or 24–hour scratch area)/0–hour scratch area × 100%.

Transwell migration and invasion

RBE and HCCC9810 cells were counted at 3 × 10⁴ cells for the transwell migration assay and 5 × 10⁴ cells for the transwell Matrigel invasion assay. The cells were resuspended in 100 μL of serum–free medium, thoroughly mixed, and then added to the upper chamber of the transwell. The lower chamber was filled with 600 μL of complete medium. The cells were then cultured for 24 h and 48 h, after which the cells on the lower surface of the membrane were fixed with 4% paraformaldehyde. The fixed cells were subsequently stained with crystal violet. Images were captured using an optical microscope (Olympus, Japan). Finally, the number of migrated and invaded cells was measured via ImageJ software.

Colony formation assay

The cells in good condition were digested and resuspended, and 1,000 cells per well were evenly seeded into a 6–well plate. The culture medium was replaced every three days. After 14 days of culture or when visible clones formed, 1 mL of 4% paraformaldehyde was added for fixation. Following crystal violet staining, the number of clones formed was quantitatively analyzed using ImageJ software.

Cell cycle analysis

The cells were digested with trypsin, rinsed once with precooled PBS, and centrifuged at 1,000 g for 5 minutes. The cell pellet was mixed with precooled 70% ethanol, fixed at 4 °C for 2 hours or overnight, and then centrifuged at 1,000 g for 5 minutes to pellet the cells. The cells were resuspended in PBS again and centrifuged again to pellet the cells, 0.5 mL of prepared PI staining solution was added (Yeason, China), the mixture was mixed well to resuspend the cells, the mixture was incubated at 37 °C in the dark for 30 minutes, and the cell cycle distribution was analyzed again via a flow cytometer (Thermo Fisher, USA).

Animal experiments

Four–week–old male BALB/c nude mice were obtained from SPF (Suzhou) Biotechology Co., Ltd. and housed under pathogen–free conditions, following the institutional animal care guidelines. All animal experiments were approved by the Ethics Committee of Fujian Cancer Hospital. All animal experiments, which included euthanasia, were conducted in accordance with AAALAC and IACUC standards. To create a subcutaneous xenograft model, 5 × 10⁶ cells in 100 μL of PBS were injected into the axillary region of the mice. After 4 weeks, the mice were sacrificed, and the tumors were surgically removed to measure their volume and weight.

RNA-seq

In accordance with the manufacturer’s protocol, total RNA was extracted from the knockdown and control cells via TRIzol reagent. The mRNAseq libraries were subsequently constructed from the total RNA via the Illumina TruSeq mRNA Sample Preparation Kit and sequenced on the Illumina HiSeq 2,500 platform with a 2 × 150 bp setting. The assembled transcripts were then compared with the National Center for Biotechnology Information (NCBI) database. Gene expression levels were quantified via Htseq software (V0.6.1), which employs the fragments per kilobase of transcript per million mapped reads (FPKM) method, to calculate gene expression.⁷⁴ Differential expression analysis was performed via the DESeq2 R package in R (v4.3.1). We have fully uploaded the raw RNA-seq sequencing files in fastq.gz format to the SRA database (PRJNA1336329).

ATAC-seq

A total of 50,000 cells were centrifuged at 4 °C for 5 min at 500 g, after which the supernatant was removed. The cells were washed once with cold PBS. After centrifugation at 500g for 5 min at 4 °C, the supernatant was removed. The cells were subsequently suspended in cold lysis buffer. After centrifugation at 500g at 4 °C for 10 min, the supernatant was removed. The transposing reaction system was configured with the Tn5 transposase. The cell nuclei were suspended with the transposing reaction system, and the DNA was purified after incubation at 37 °C for 30 min. The PCR system was configured with the purified DNA, and then the PCR amplification reaction was performed. The final DNA libraries were run on an Illumina platform after the DNA was purified.⁷⁵

Dual-luciferase reporter assay

Wild–type and mutant HERV2365 and FGF1 fragments were constructed and inserted downstream of the luciferase reporter gene in the reporter plasmid psicheck2.0 (Zolgene, China). 293T cells were seeded in 6–well plates and cultured to reach 60–70% confluence within 24 h. Subsequently, the reporter plasmids were transfected into the cells using Lipofectamine 3000. Cells were cotransfected with different combinations of plasmids harboring the 3′–untranslated region (3′–UTR) of assayed genes and miRNA mimics or the negative control. After 48 h, the activities of both firefly luciferase and Renilla luciferase were measured with a Dual–Luciferase Reporter System Kit (Vazyme, China). The detailed sequence information for the wild-type and mutant binding sites of HERV2365 and FGF1 is provided in Data S1A and S1B.

Fluorescence in situ hybridization (FISH)

The FISH probe mix for HERV2365 was synthesized by Servicebio (Wuhan, China), and FISH assays were performed via the Fluorescent In Situ Hybridization Kit (Servicebio, China). RBE and HCCC9810 cells were fixed with situ hybridization fixative (animal) and then incubated with the probe overnight at 37 °C. Nuclei were stained with DAPI, the sections were observed under a Nikon upright fluorescence microscope, and images were collected.

Western blot

Proteins were extracted with RIPA buffer (Beyotime) containing PMSF. Equal amounts of protein were separated by SDS‒PAGE, transferred to PVDF membranes, and probed with anti–FGF1 (Abcam, ab207321) and anti–GAPDH (Proteintech, 60004–1–Ig) antibodies. The signals were detected via WesternBright ECL (Advansta, USA).

Coding potential analysis

Coding potential was evaluated using CPAT, CNCI, PLEK, and CPC2 with default settings. Candidate ORFs were identified and the top ORF was translated into an amino acid sequence for protein-level annotation using InterProScan with comprehensive member databases (including Pfam, SMART, CDD, PANTHER, SUPERFAMILY, and CATH-Gene3D). Protein homology was further assessed by BLASTP against the curated RefSeq Select protein database (refseq_select_prot) using default parameters.

Quantification and statistical analysis

Statistical analyses were performed with GraphPad Prism (v9.3.5) and R (v4.3.1). Comparisons between groups were made using Student’s t–test or one–way ANOVA. Survival curves were estimated using the Kaplan–Meier method, and differences were assessed with the log-rank test. Cox proportional hazards regression was used for prognostic evaluation. Correlations were assessed by Pearson’s test. Each experiment was independently repeated three times. Differences were considered statistically significant when ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, or ∗∗∗∗p < 0.0001.

Published: April 22, 2026